cian911/smart-gas-meter

smart-gas-meter is a Python project that automates the reading of a physical gas meter using a Raspberry Pi and a camera. Every 10 minutes the Pi takes a photo of the meter's LCD display. A computer vision model then identifies and reads the digits from the photo, reconstructs the full meter reading, and sends it to Home Assistant (a popular home automation platform). There is also an optional script that submits readings directly to Gas Networks Ireland's website. The digit-recognition model is a fine-tuned version of YOLOv8n, a compact object-detection model. It was trained on over 3,000 meter photos and achieves near-perfect accuracy. Because it detects and locates digits anywhere in the image rather than looking in fixed positions, it handles small shifts in the camera's position automatically without needing recalibration each time. A pre-trained model is included in the repository so you can start testing with photos immediately. If your meter looks different from the one used in training, the repository includes all the scripts needed to build your own labeled dataset and retrain the model. You calibrate a region of interest by drawing a box around the display, label a batch of photos by typing the reading you see, convert those labels into the training format, and retrain. Fine-tuning takes roughly an hour on a laptop with a GPU. The trained model is then exported to ONNX format (a lightweight format that works without the full PyTorch library) for deployment on the Pi. Home Assistant integration creates a sensor called sensor.gas_meter that appears as a gas consumption source in the Energy dashboard. The Gas Networks Ireland submission script handles the multi-step web form process automatically, including fetching a one-time security token before posting the reading. The repository does not specify a license.